EP0696404B1 - Differential coding and decoding method and related circuits - Google Patents

Abstract

PCT No. PCT/EP94/01360 Sec. 371 Date Dec. 27, 1995 Sec. 102(e) Date Dec. 27, 1995 PCT Filed Apr. 29, 1994 PCT Pub. No. WO94/26068 PCT Pub. Date Nov. 10, 1994A method for the differential coding and decoding and related circuits wherein the differential coding is a special type of coding which, instead of coding directly an input signal, carries out the difference between this signal and a predicted signal of it. This allows the reduction of the information to be transmitted. The method is capable of further reducing the signal redundancy and therefore the information to be transmitted.

Description

• The present invention relates to a method and circuits for differential coding and decoding.
  A method normally used for reducing the information to be transmitted is to carry out the difference between the signal in transit and its predicted one.
  At this point the problem arises in that if the input signal is confined within a certain range of values, then the signal, obtained by carrying out the difference between input signal and predicted signal, has a double range.
  Simple algorithms, capable of maintaining the differential signal within the range of the original signal, are known.
• For example the book by W. K. Pratt: Digital Image Processing, 1978, John Wiley & Sons, US; Section 22.5.3: "Differential Pulse Code Modulation" on pages 641-645, describe a method for coding a video signal where the difference between an actual pixel and its estimate is quantized and coded for transmission. Further by the fact that small differences are more numerous than large differences, it is possible to employ a variable length code and achieve a greater coding compression.
• It is also known from the article of Electronics & Communications in Japan, Part 1, vol. 73, n. 6, June 1990, pages 12-21; Y. Izawa et al.: "Improvement of picture coding and picture efficiency using Discrete Cosine Transform", an improvement of the picture quality and coding efficiency using discrete cosine transform.
• It is an object of the present invention to provide a method of differential coding and decoding which is capable of reducing efficiently the redundancy of the information to be transmitted. In accordance with the invention the coding method is therefore structured as set forth in claim 1, and the related circuit as set forth in claim 3.
  The invention will become more intelligible from the following description.
• The differential coding is a special type of coding which, instead of coding directly an input signal, carries out the difference between this signal and a predicted signal of it. For the prediction of the input signal it is possible to use any difference between this signal and a predicted signal of it. For the prediction of the input signal it is possible to use any type of predictor; in the simplest case the value encoded previously can be used as prediction signal.
  Assuming that the input signal values are comprised in the set (-range, ......, range -1), range being a positive integer, it results that the difference between this signal and its predicted one has a double range.
  Therefore, the problems arises as to how the differential signal can be led within the original range.
  The algorithm herein set forth is known from the literature and is the simplest method for maintaining the differential signal within the original range signal.
  Let:

val

be the value of the sample to be predicted

pred

be the prediction value.

• The algorithm is the following:
     Δ = val - pred;
     if ( Δ < - range ) Δ = Δ + 2 range;
     else if ( Δ > range - 1 ) Δ = Δ - 2 range;
• In an embodiment thereof, the invention, besides maintaining the values of the differential signal within the original range, carries out a further compression thereof.
  The underlying idea is that, given the predicted signal, the difference values are allocated in such a way as to have the smaller values as close as possible to the prediction value.
  In this manner, more probable values are assigned smaller numbers.
  The algorithm is the following:
  Δ = val - pred;
  range _Δ = range ;
  if ( pred > 0 ) range _Δ = range - pred - 1;
  else if ( pred < 0 ) range _Δ = range + pred;
  if ( ¦Δ¦ > range _Δ )
     if (( ¦Δ¦ - range _Δ + 1 ) mod 2 == 0 ) sn = -1;
     else if (( ¦Δ¦ range _Δ + 1 ) mod 2 == 1 ) sn = + 1;
     Δ = ( range _Δ + ( ¦Δ¦ - range _Δ + 1 ) / 2 ) sn;
• When the input signal is zero for a great number of times, the difference operation with the previous one tends to eliminate some of the zeroes which are present.
  It is convenient to maintain the sequences of zeroes, since in an eventual subsequent compression of information, they can be encoded in a very effective manner.
  In another embodiment of the invention, by carrying out an absolute coding of the signal when its value is zero, the number of zeroes remain unchanged even working on differential.
  The algorithm is the following:
  if ( val== 0 ) Δ = 0;
  else if ( pred == 0 ) Δ = val;
  else if ( pred > 0 )
     Δ = val - pred;
     if (( Δ ≥ 0 ) ∨ ( Δ < - pred )) Δ = Δ + 1;
  else if ( pred < 0 )
     Δ = val - pred;
     if (( Δ ≤ 0 ) ∨ ( Δ > - pred )) Δ = Δ - 1;
• No control on differential signal range has been introduced here; however the techniques of the algorithms described previously can be extended thereto.
  An application to video signal of the above-mentioned method will now be described.
  The fundamental block of the system is the predictor that allows the variation range of the higher energy transformed coefficients to be reduced. The differential signal which is transmitted has a statistic feature that is very convenient for a subsequent coding.
  The method is applied for the coding of the video image components, luminance and chrominance, in which one resorts to DCT (Discrete Cosine Transform).
  A digital image is composed of a number of lines and each line is formed by a number of dots commonly referred to as "pixels".
  The video coding technique to which reference is made herein, carries out a decomposition of the image into blocks of 8x8 pixels and then it applies the bidimensional DCT to each block. Such operation allows the spatial redundancy of the pixels in each block to be reduced.
  At this point the DCT coefficients inside each block are uncorrelated each other; it remains to be seen the residual correlation between adjacent blocks.
  In the method of the invention blocks along the strips of the image are considered. In particular the first block of each strip is transmitted without any processing, starting from the next one the method of the invention is applied in order to reduce the values of the higher weight coefficients.
  The algorithm considers only the coefficients of the first column of each transformed 8x8 block; such coefficients generally are those of higher weight since they represent the vertical frequencies that, in an interlaced scanning, are the most important ones.
  Therefore, just as one is processing the present block, he knows all the preceding block and all the coefficients of the present block with the exception of those of the first column, that are to be estimated.
  The method consists in linking up the waveform (in the space domain) to the border between the present block and the preceding one.
  To do so the preceding block is antitransformed by columns and the last column is determined; also the present block is antitransformed by columns to determine the first column, but in this circumstance the DCT coefficients of the first column are not known and therefore appear as unknowns in the antitransform.
  At this point the continuity between the two blocks is imposed by linking up the two columns at the common border (in the space domain).
  From the system so obtained the unknown terms of the first column are obtained.
  All this procedure consists merely in the following calculation: $$\begin{gathered} {\mathit{\text{PR}}}_{\mathit{\text{x0}}}\text{=}{\mathit{\text{C}}}_{\mathit{\text{x0}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x1}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x1}}}^{\mathit{\text{att}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x2}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x2}}}^{\mathit{\text{att}}} \\ \text{-}{\mathit{\text{C}}}_{\mathit{\text{x3}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x3}}}^{\mathit{\text{att}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x4}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x4}}}^{\mathit{\text{att}}} \\ \text{-}{\mathit{\text{C}}}_{\mathit{\text{x5}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x5}}}^{\mathit{\text{att}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x6}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x6}}}^{\mathit{\text{att}}} \\ \text{-}{\mathit{\text{C}}}_{\mathit{\text{x7}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x7}}}^{\mathit{\text{att}}} \end{gathered}$$

  

     x = 0,..., 7
  where:
  PR_x0  is the predicted value of the DCT coefficients with index x0 of the present block,
  C $\genfrac{}{}{0ex}{}{\text{pre}}{\text{xy}}$

  

    is the DCT coefficient with index x, y of the preceding block;
  C $\genfrac{}{}{0ex}{}{\text{att}}{\text{xy}}$

  

    is the DCT coefficient with index x, y of the present block.
• The prediction thus obtained is subtracted from the true value and the difference is transmitted.
  It is to be noted that the value estimated through the foregoing formula could assume also values which are out of the allowed range; in this circumstance a saturation at the closest end is merely carried out.
  Next step consists in introducing suitable weight coefficients for each pair of homologous coefficients. In this manner it is considered that it is not correct to link up the columns at the borders, particularly if the image varies too much from one block to another. Such correction is indispensable especially for higher order coefficients. In this case the formula becomes: $$\begin{gathered} {\mathit{\text{PR}}}_{\mathit{\text{x0}}}\text{=}{\mathit{\text{K}}}_{\text{0}}{\mathit{\text{C}}}_{\mathit{\text{x0}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{K}}}_{\text{1}}{\mathit{\text{(C}}}_{\mathit{\text{x1}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x1}}}^{\mathit{\text{att}}}\mathit{\text{)}}\text{+}{\mathit{\text{K}}}_{\text{2}}{\mathit{\text{(C}}}_{\mathit{\text{x2}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x2}}}^{\mathit{\text{att}}}\mathit{\text{)}} \\ \text{-}{\mathit{\text{k}}}_{\text{3}}{\mathit{\text{(C}}}_{\mathit{\text{x3}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x3}}}^{\mathit{\text{att}}}\mathit{\text{)}}\text{+}{\mathit{\text{K}}}_{\text{4}}{\mathit{\text{(C}}}_{\mathit{\text{x4}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x4}}}^{\mathit{\text{att}}}\mathit{\text{)}} \\ \text{-}{\mathit{\text{k}}}_{\text{5}}{\mathit{\text{(C}}}_{\mathit{\text{x5}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x5}}}^{\mathit{\text{att}}}\mathit{\text{)}}\text{+}{\mathit{\text{K}}}_{\text{6}}{\mathit{\text{(C}}}_{\mathit{\text{x6}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x6}}}^{\mathit{\text{att}}}\mathit{\text{)}} \\ \text{-}{\mathit{\text{K}}}_{\text{7}}{\mathit{\text{(C}}}_{\mathit{\text{x7}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x7}}}^{\mathit{\text{att}}}\mathit{\text{)}} \end{gathered}$$

  

     x = 0,...,7
  where:
     K _x = weight coefficient.
• From simulations carried out it has been found that the performances of the system are not very sensitive to variation of weight coefficients, whereby they have been chosen all equal and with values given by sums of base two powers in order to simplify the relative hardware.
  In this circumstance the estimated value in general occupies a number of bits greater than the allowed one, a simple uniform quantization with saturation leads the signal again in the correct range.
  The prediction value is subtracted from the true value by following two different algorithms, one if the DCT coefficient has a 00 index and one for the other coefficients.
  When DCT coefficient's index is 00, the first described algorithm is used, which allows the prediction error range to be reduced. In fact, such coefficient has a quasi-uniform Probability distribution; the difference operation with the predicted one transforms such distribution into a Laplace distribution which is much more convenient for a successive coding.
  For the other coefficients to be predicted, since they are zeroed for a great number of times, the algorithm of the differential with keeping of zeroes is used.
  In case of higher order coefficients, the prediction so carried out does not allow the achievement of a significant compression of the signal; in such circumstance the coefficient is transmitted without any processing.
  As far as the differential signal decoding is concerned, it is sufficient to carry out the inverse operations with respect to those described in the coding algorithms.
  The circuits for implementing said methods in accordance with the invention comprise at least a predictor, an adder or subtracter and means capable of carrying out calculations like, e.g., a digital signal processor.
  In the simplest case, the predictor can be represented like a delay line and output the value transmitted previously. Obviously several variations of the methods and circuits as described above are possible all falling anyway within the scope of the present invention.

Claims (3)

1. A method for coding a video signal comprising the steps of:

   - considering blocks of X rows and Y columns of said video signal;

   - calculating the bidimensional Discrete Cosine Transform of said blocks;

   - predicting the transform coefficients of the first column of each transformed block;

   - carrying out a difference between the true value of each of said transform coefficients and the predicted value of each of said transform coefficients if said true value is different from zero; and characterised by

   - predicting said transform coefficients of the first column of each transformed block using coefficients of further columns of said transformed block and coefficients of a preceding transformed block;

   - coding said difference by associating more probable difference values to smaller numbers, thereby maintaining said numbers within the initial range of said true value; and

   - coding the true value of said transform coefficient as such if said true value is equal to 0.
2. A method according to claim 1 characterised in that said predicted first column of each block have been calculated by means of the following calculation: $$\begin{gathered} {\mathit{\text{PR}}}_{\mathit{\text{x0}}}\text{=}{\mathit{\text{K}}}_{\text{0}}{\mathit{\text{C}}}_{\mathit{\text{x0}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{K}}}_{\text{1}}{\mathit{\text{(C}}}_{\mathit{\text{x1}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x1}}}^{\mathit{\text{att}}}\mathit{\text{)}}\text{+}{\mathit{\text{K}}}_{\text{2}}{\mathit{\text{(C}}}_{\mathit{\text{x2}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x2}}}^{\mathit{\text{att}}}\mathit{\text{)}} \\ \text{-}{\mathit{\text{K}}}_{\text{3}}{\mathit{\text{(C}}}_{\mathit{\text{x3}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x3}}}^{\mathit{\text{att}}}\mathit{\text{)}}\text{+}{\mathit{\text{K}}}_{\text{4}}{\mathit{\text{(C}}}_{\mathit{\text{x4}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x4}}}^{\mathit{\text{att}}}\mathit{\text{)}} \\ \text{-}{\mathit{\text{K}}}_{\text{5}}{\mathit{\text{(C}}}_{\mathit{\text{x5}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x5}}}^{\mathit{\text{att}}}\mathit{\text{)}}\text{+}{\mathit{\text{K}}}_{\text{6}}{\mathit{\text{(C}}}_{\mathit{\text{x6}}}^{\mathit{\text{pre}}}\text{-}{\mathit{\text{C}}}_{\mathit{\text{x6}}}^{\mathit{\text{att}}}\mathit{\text{)}} \\ \text{-}{\mathit{\text{K}}}_{\text{7}}{\mathit{\text{(C}}}_{\mathit{\text{x7}}}^{\mathit{\text{pre}}}\text{+}{\mathit{\text{C}}}_{\mathit{\text{x7}}}^{\mathit{\text{att}}}\mathit{\text{)}} \end{gathered}$$

   

   x = 0,...,7 where:

   PR _x0 =   is the predicted value of the DCT coefficients with index x,0 of the present block

   C $\genfrac{}{}{0ex}{}{\mathit{\text{pre}}}{\mathit{\text{xy}}}$

   

      = is the DCT coefficient with index x,y of the preceding block

   C $\genfrac{}{}{0ex}{}{\mathit{\text{att}}}{\mathit{\text{xy}}}$

   

      = is the DCT coefficient with index x,y of the present block
      K _x = weight coefficient
3. A circuit for coding a video signal comprising:

   - means for considering blocks of X rows and Y columns of said video signal;

   - means for calculating the bidimensional Discrete Cosine Transform of said blocks;

   - means for predicting the transform coefficients of the first column of each transformed block;

   - means for carrying out a difference between the true value of each of said transform coefficients and the predicted value of each of said transform coefficients if said true value is different from zero;

   characterised in that

   - said means for predicting the transform coefficients of the first column of each transformed block uses coefficients of further columns of said transformed block and coefficients of a preceding transformed block; and said circuit further comprises

   - means for coding said difference by associating more probable difference values to smaller numbers, thereby maintaining said numbers within the initial range of said true value; and

   - means for coding the true value of said transform coefficient as such if said true value is equal to 0.